What can we anticipate in tech for 2025? Already, we’ve seen vast innovations coming out in 2024 and these buzzwords were mentioned practically everywhere, at our workplaces and conferences, in news headlines and our social media feeds. Some were groundbreaking while others were hype or sketch at best.
The 2024 trendy tech buzzwords were:
- Virtual Reality (VR) and Augmented Reality (AR),
- The Metaverse
- Quantum Computing
- Artificial Intelligence (AI) and Machine Learning (ML)
- Generative AI (Gen AI) and Large Language Models (LLMs)
- Crypto and Blockchain (thank goodness NFTs “died” in 2024).
- Cloud Computing
We’ve seen how organisations have used VR/AR to offer groundbreaking experiences in various domains (e.g., BMW using AR to prototype vehicles in their plants). We’ve also seen that cloud technology still dominated the tech landscape, with introduction of various development tools (code assistance), security services and AI services from each cloud computing service companies (Amazon Q AI Agent and Microsoft Copilot AI Service on Azure and inclusion for LLMs such as LLama3 on both Amazon Bedrock and Microsoft Azure).
Serverless seems to be a popular cloud computing service choice due to its a pay-as-you-go model, offering scalability and cost-efficiency by charging only for the resources used. Moreover, it enables faster development cycles, enhances agility, and supports the creation of microservices-based architectures.
So, as we’ve entered 2025, here’s my prediction of what we will see trending this year.
AI: The year of intelligence
For the past few years Generative AI has inculcated the world. It has impacted the way we work, consume content and, possibly live. our lives. And that’s not the end. The future of tech will be AI infused (in hardware, software and business services) that AI wil be ubiquitous, just the way that the HTTP is for the internet.
These are the trends to watch that will render AI ubiquitous:
- Agentic AI: At the last quarter of 2024 we’ve seen the emergence of agentic AI. Unlike previous AI agents, which were rule-based and required human intervention, Agentic AI are “proactive” agents that autonomously solve complex, multi-step problems (they get the context of the problem) and achieve goals without human intervention by using sophisticated reasoning (see below) and iterative planning. Tesla’s Full Self-Driving system is an example of Agentic AI that continuously learns from the driving environment and adjusts its behavior to improve safety and efficiency. We’re expecting to see more agentic AI in the various field, like Cybersecurity (analysis potential threats and automatically respond to such threats), Healthcare (e.g., patient care management), IT DevOps (e.g., production log monitoring to detect production bugs, write potential bugfixes, test them and create a pull-request with a patch of the potential fix), etc.
- Reasoning/Thinking Large Language Models: Most LLMs generate a response by generally predict the next token to generate at each step of its transformer architecture, and this has proven to be reasonably effictive but it is also proned to produce inaccurate response or fall into hallucinations. Recently, reasoning has been an interesting research in the foundational models since the release of GPT-o1 by OpenAI. Meta AI’s recent research paper (published in October 2024) on Thinking LLMs with Thought Generation - Thought Preference Optimization (TPO) - is one step closer to answering the question on how human experts responds to questions.
Large Language Models with reasoning, also known as Large Reasoning Models (LRM) will help solve complex problems or derive to new knowledge by creating new hypothesis from existing knowledge base and creating tests to validate its hypothesis, at the same time providing and outlining reasoning steps the AI arrived to its “conclusion”.
We’re expecting more reasoning & fine-tuned AI models. Google DeepMind released Google Gemini 2.0 Flash Thinking Mode (a prototype) “best for multimodal understanding, reasoning, and coding,” (their words), which you can try it out at their Google AI Studio. - Small language models: As more tech firms are building AI chips, and more devices will be built with AI in mind, we will see more fine-tuned small Large Language Models, with few billions of parameters in size, that can operate on-device with very little compute overhead specialised to do specific tasks (through agent AIs).
- More AI specialised hardware and computing services: As more AI chips, AI computing hardwares, AI datacenters continue to evolve and expand the more we’ll see vast amount of AI startups, AI services and AI product integration offerings for organisations to easily create bespoke AI agents and AI solutions for their bespoke needs. Again, this speaks to the ubiquitous nature AI will have to the tech ecosystem and the world at large.
- Very Large Language Models: With refined training parameters and larger parameter sizes, in their trillions. OpenAI release GPT-4 with over 1 trillion parameters, Baidu’s Wu Da 2.0 has 1.55 trillion parameters, and Amazon Nova LLM with a staggering 2 trillion parameters. These are intended to provide compelling intelligence, better content generation while delivering better capabilities such as reasoning tasks and agentic capabilities.
While we can carry on the rabbit hole of listing other predictions (such as infinite context window size - to memorise everything conversation you’ve had with the chatbot, inference compute, etc.), the biggest trend, undoubtedly, will have to come from organisations with the goal aimed towards customer experience, powered by AI. Organisations will have to provide Intelligence to their existing offerings:
- AI-powered B2B solutions, such as rapid product development and innovation, predictive analytics and insights, customer support, process automations (streamlining complex workflows, improve efficiency and human errors) and tailored content generation for businesses.
- AI-powered B2C solutions, such as personalised customer services (chatbots that provides immediate feedback to customer requests/inquiries), personalised marketing (e.g., targeted tailored AI ads based on user behaviours), etc.
We will still see innovation and advancement in the field of quantum computing. Google has joined the quantum chip era by releasing the Willow chip in December 2024, and cloud providers are also upgrading their quantum cloud offerings. Amazon Braket (the quantum computing service of Amazon Web Services (AWS)) onboarding the latest superconducting quantum processor from Rigetti Computing, the 84-qubit Ankaa-2 device while Microsoft has announced that 2025 is the year to be Quantum-ready.
Now, let me ask you, what are the tech trends that you foresee coming in 2025? Let’s have this discussion by commenting below.
Happy 2025! 🎉
Comments
Post a Comment